Architecture of retinal projections to the centralcircadian pacemakerDiego Carlos Fernandeza,1, Yi-Ting Changb,1, Samer Hattara,c, and Shih-Kuo Chenb,d,2

aDepartment of Biology, Johns Hopkins University, Baltimore, MD 21218; bDepartment of Life Science, National Taiwan University, Taipei 10617, Taiwan;cDepartment of Neuroscience, Johns Hopkins University, Baltimore, MD 21218; and dGenome and Systems Biology Degree Program, National TaiwanUniversity and Academia Sinica, Taipei 10617, Taiwan

Edited by Joseph S. Takahashi, Howard Hughes Medical Institute, University of Texas Southwestern Medical Center, Dallas, TX, and approved April 13, 2016(received for review November 30, 2015)

The suprachiasmatic nucleus (SCN) receives direct retinal input fromthe intrinsically photosensitive retinal ganglion cells (ipRGCs) forcircadian photoentrainment. Interestingly, the SCN is the only brainregion that receives equal inputs from the left and right eyes.Despite morphological assessments showing that axonal fibersoriginating from ipRGCs cover the entire SCN, physiological evi-dence suggests that only vasoactive intestinal polypeptide (VIP)/gastrin-releasing peptide (GRP) cells located ventrally in the SCNreceive retinal input. It is still unclear, therefore, which subpopula-tion of SCN neurons receives synaptic input from the retina and howthe SCN receives equal inputs from both eyes. Here, using singleipRGC axonal tracing and a confocal microscopic analysis in mice, weshow that ipRGCs have elaborate innervation patterns throughoutthe entire SCN. Unlike conventional retinal ganglion cells (RGCs) thatinnervate visual targets either ipsilaterally or contralaterally, asingle ipRGC can bilaterally innervate the SCN. ipRGCs form synapticcontacts with major peptidergic cells of the SCN, including VIP, GRP,and arginine vasopressin (AVP) neurons, with each ipRGC innervat-ing specific subdomains of the SCN. Furthermore, a single SCN-projecting ipRGC can send collateral inputs to many other brainregions. However, the size and complexity of the axonal arboriza-tions in non-SCN regions are less elaborate than those in the SCN.Our results provide a better understanding of how retinal neuronsconnect to the central circadian pacemaker to synchronize endoge-nous circadian clocks with the solar day.

melanopsin | circadian | suprachiasmatic nucleus |non-image–forming functions | ipRGCs

The suprachiasmatic nucleus (SCN) houses a central pacemakerthat orchestrates circadian (circa: “about” and diem: “day”)

behaviors that cycle with a period close to 24 h. This endogenousclock is self-sustained even in the complete absence of externalstimuli, but can be entrained to the light/dark cycle of the solar dayin a process defined as circadian photoentrainment (1). In mam-mals, phototransduction solely occurs in the retina (2). Retinalganglion cells (RGCs) are projection neurons of the retina thatsend light information to brain targets (3). Although the majorityof RGCs project to brain areas involved in object tracking andimage formation, some RGCs also innervate non-image–formingbrain regions to influence circadian activity, sleep, the pupillarylight response, mood, and learning functions (4).Recent studies showed that a small subpopulation of RGCs

expresses a photopigment called melanopsin (Opn4), makingthese cells intrinsically photosensitive (ip)RGCs (5–7). It is nowknown that there are at least five different ipRGC subtypes (M1–M5) (8, 9). Using genetic tracing techniques, it was shown thatipRGCs send projections to non-image–forming centers in thebrain, including the SCN, which receives retinal input predominantlyfrom M1 ipRGCs (10–12) and to a lesser extent from M2 ipRGCs(12). The SCN is a small nucleus located in the anterior hypo-thalamus and contains diverse subpopulations of neurons (13).Studies in rodents indicate that the core region located in theventral portion of the SCN is composed of neurons that expressvasoactive intestinal polypeptide (VIP) or gastrin-releasing peptide

(GRP) (14, 15). This core region is surrounded by a shell regionthat contains neurons expressing arginine vasopressin (AVP) andcalretinin (16). To deconstruct the SCN circuitry, the rhythmicexpression of clock genes and the photic induction of immediateearly genes have been evaluated in different subpopulations ofSCN neurons. Those studies gave rise to a functional description ofthe SCN, in which neurons in the shell region are classified ascircadian oscillators, whereas neurons in the ventral region arelight-responsive (17). In rats, it was shown that specific cell types inthe core region, such as VIP- and GRP-expressing cells, receivedirect retinal input (18–20), which led to the idea that the coreregion of the SCN is the only retinorecipient area. However, thoseresults were inconsistent between different rodent species, as noretinal-VIP–positive (VIP+) cell contacts were observed in ham-sters (21). Furthermore, this simple demarcation of the SCN waschallenged by studies showing that the entire mouse SCN receivesdense innervation from ipRGCs (11). Because of the dense retinalinnervation pattern, it has been difficult to decipher the identity ofSCN neurons innervated by ipRGCs (22–24). Therefore, little isknown about which cell types in the SCN receive retinal input. Inaddition, the SCN is the only region in the mouse brain that re-ceives nearly equal inputs from the ipsilateral and contralateraleyes. This is quite surprising because all other retinorecipient brainregions are innervated ∼95% contralaterally (23). The importanceof and the mechanism of how the SCN receives equal input fromboth eyes are essentially unexplored.Here, we used single ipRGC tracing and confocal microscopic

methods to show that ipRGCs have elaborate patterns of SCNinnervation and found that several kinds of peptidergic neurons,

Significance

Most physiological processes exhibit circadian oscillations, whichare synchronized by a central pacemaker located in the hypo-thalamic suprachiasmatic nucleus (SCN). For this pacemaker to bebiologically relevant, it must be entrained with external envi-ronmental cues such as the daily light/dark cycle. At present,details of how photic information is relayed from the retina tothe SCN remain unclear. Using an array of genetic mouse lines,we found that the major peptidergic SCN neurons receive directretinal input and that a single intrinsically photosensitive retinalganglion cell (ipRGC) bilaterally targets the SCN and sends axo-nal collaterals to several non-SCN regions. Together, our resultssuggest that the retina provides multifaceted synaptic inputs tothe brain to mediate proper photic inputs to coordinately influ-ence non-image–forming visual functions.

Author contributions: D.C.F., Y.-T.C., S.H., and S.-K.C. designed research; D.C.F. and Y.-T.C.performed research; D.C.F., Y.-T.C., and S.-K.C. analyzed data; and D.C.F., Y.-T.C., S.H., andS.-K.C. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1D.C.F. and Y.-T.C. contributed equally to this work.2To whom correspondence should be addressed. Email: alenskchen@ntu.edu.tw.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1523629113/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1523629113 PNAS | May 24, 2016 | vol. 113 | no. 21 | 6047–6052

NEU

ROSC

IENCE

located in topographically distinct areas of the SCN, receive directretinal input. In addition, we uncovered the mechanism of how theSCN receives equal inputs from both eyes and showed that a singleipRGC sends axonal collateral projections to multiple brain re-gions. Together, our studies provide strong anatomical evidencefor how ipRGCs project to the brain to influence light-mediatedbehaviors.

ResultsA Single M1 ipRGC Projects to Multiple Brain Nuclei. To reveal theaxonal terminal fields of a single ipRGC, we crossed anOpn4CreERT2

mouse line with a Z/AP reporter line to generateOpn4CreERT2/+;Z/APmice (Fig. S1). In these mice, depending on the injected concen-tration of tamoxifen, we were able to label a single ipRGC or up toseveral hundred ipRGCs with alkaline phosphatase. Specifically,with a low dose of tamoxifen, we were able to label a single ipRGC(Fig. 1A), revealing its dendritic structure in the retina and its axonalarchitecture in the brain. We further obtained a 3D reconstructionfor each single ipRGC (Fig. 1 C, F, and H). In total, we evaluated20 ipRGCs, all of which were of the M1 subtype as confirmed bytheir dendrites that stratified in the OFF sublamina of the innerplexiform layer of the retina (Fig. 1 B and C). Axonal labeling wasevident in the optic chiasm (Fig. 1D) and was traced to the brain. Alllabeled ipRGCs innervated the SCN, and as expected, fibers tar-geted the ventral SCN through the optic chiasm (Fig. 1G). A singleipRGC can also send collateral axonal projections to multiple brainnuclei, which include the intergeniculate leaflet (IGL) (Fig. 1E), theventral medial hypothalamus (VMH) (Fig. S2A), the peri-habenularregion (pHb) (Fig. S2B), the ventral division of the lateral geniculatenucleus (vLGN) (Fig. S2C), the pretectal nucleus (PN) (Fig. S2D),and/or the superior colliculus (SC) (Fig. S2E). Although an indi-vidual ipRGC has a distinct innervation pattern to various braintargets in addition to the SCN (up to five nuclei), almost all ipRGCs(19/20) sent collateral projections to the IGL. In contrast, only 1 of

20 labeled ipRGCs had axon collaterals terminating in the pHb orVMH (Fig. S2F and Table S1). Furthermore, we found that theaxonal terminal fields from ipRGCs to non-SCN targets were simpleyet variable. In the IGL, ipRGC fibers either spread throughout thethin sheet of the leaflet (Fig. S3) or made branches that innervatedonly a small region of the leaflet (Fig. 1E). In the SC, the in-nervation pattern was even simpler than that of the IGL, andipRGCs formed one or two thin straight lines that extended to deeplayers of the SC (Fig. S2E), as was previously shown (11). In con-trast, branching patterns of ipRGC fibers to the SCN were muchmore elaborate (Fig. 1 G and H). Specifically, retinal fibers withinthe SCN were tortuous and covered a large volume of the nucleuswith prominent terminal swellings (Fig. S4). In fact, the estimatedvolume of the axonal terminal (axon volume 3.4 × 106 ± 0.60 ×106 μm3) coverage from an individual ipRGC was a remarkable∼14.4% of the total SCN volume (SCN volume 23.6 × 106 ± 0.75 ×106 μm3) (Materials and Methods).

A Single ipRGC Can Bilaterally Innervate the SCN. The complexity ofthe retinal terminal fields in the SCN was not the only definingfactor that was specific to this brain region. Retinal innervationto the IGL, vLGN, PN, and/or SC was either ipsilateral orcontralateral with a 1:9 ratio. Innervation to the ipsilateral andcontralateral parts of the SCN from ipRGCs, however, wasnearly equal with a 4:6 ratio. We discovered that ipRGCs in-nervate the SCN in two distinct ways. First, similar to other brainregions, ipsilaterally projecting ipRGCs (2/20) only innervatedthe SCN unilaterally (Fig. 2 A and D). Ipsilaterally projectingipRGCs were located in the ventral-temporal part of the retina(Fig. 2C), similar to conventional RGCs that project ipsilaterally(25). This suggests that ipRGCs may also follow classical rules ofipsilateral targeting. Second, we found that 66.7% of con-tralaterally projecting ipRGCs (12 of 18) used atypical pathwaysby which axonal projections course through the midline to in-nervate the ipsilateral portion of the SCN (Fig. 2B) or throughbranching axons at the optic chiasm (Fig. S5). Therefore, thetotal axonal volume from ipRGCs to the ipsilateral and contra-lateral parts of the SCN is close to 4:6 (Fig. 2D). Together, theseclassical and atypical ipsilateral projection patterns to the SCNexplain why the SCN is equally innervated, as published in pre-vious reports (11, 16, 26).To evaluate whether the atypically projecting ipRGCs consti-

tute a unique retinal neuronal subtype with a distinct dendriticmorphology, we analyzed and compared somato-dendritic mor-phological characteristics of ipRGCs based on their bilateral orunilateral features. We found no significant differences in any

dLGN

vLGN

IGL

A

D

B

E

G

Reconstructioncoronal sagittal

SC

N

AP stainingcoronal

Reconstructioncoronal sagittal

AP stainingwhole mount

IGL

Opt

ic x

R

etin

a

AP stainingcoronal

Reconstructionwhole mount section view

AP stainingwhole mount

F

ox

3vSCN

C

H

D

VNT D

VR C

Fig. 1. Dendritic and axonal reconstruction of a single M1 ipRGC. Represen-tative images of alkaline phosphatase staining and 3D reconstruction from asingle ipRGC are shown. A singleM1 ipRGC was labeled in the left retina (A–C);its axon crossed the midline in the optic chiasm (D), extensively innervated thecontralateral side of the SCN (G and H), and collaterally projected to the IGL(E and F). C, caudal; D, dorsal; dLGN, dorsal part of the lateral geniculatenucleus; N, nasal; ox, optic chiasm; 3v, third ventricle; V, ventral; vLGN, ventralpart of the lateral geniculate nucleus; R, rostral; T, temporal. The arrowheadsand red trace indicate the axon. Dashed lines are estimations of the boundaryof the SCN and IGL. [Scale bars: 100 μm (A); 50 μm (B–G).]

contralateralipsilateral

innervation in con SCNinnervation in ipsi SCN

ContralateralipRGCs

Unilateral

IpsilateralipRGCs

TotalipRGCs

Unilateral

Bilateral

n= 18 + 2 = 20

Unilateral

Bilateral

24.3%

46.8%

28.9%

7.7% 22.5%

43.2%

26.6%

D

V

N T

AP stainingcoronal

Reconstructioncoronal sagittal

AP stainingcoronal

3v

ox

SCN

3v

oxSCN

Reconstructioncoronal sagittal A B

C D

Fig. 2. Bilateral innervation of a single ipRGC contributes to the symmetricalinput to the SCN. (A and B) Representative images showing a single unilateralipRGC innervating the ipsilateral side of the SCN (A) and a bilateral ipRGC in-nervating both contralateral and ipsilateral sides of the SCN (B). (C) Spatial distri-bution of cell bodies in the retina for ipsilaterally (closed circles) and contralaterally(open circles) projecting ipRGCs. (D) Pie charts of innervation percentages in thecontralateral and ipsilateral SCN from ipRGCs. n = 20. (Scale bar: 50 μm.)

6048 | www.pnas.org/cgi/doi/10.1073/pnas.1523629113 Fernandez et al.

somato-dendritic morphological property between bilaterallyand unilaterally projecting ipRGCs (Fig. S6).

Individual ipRGCs Project to Specific SCN Regions. We then in-vestigated topographical patterns of ipRGC projections to the SCN.ipRGC axonal fields innervated all regions of the SCN, includingthe ventral (Fig. 3A and Movie S1), medial (Fig. 3B and Movie S2),and dorsal (Fig. 3C and Movie S3) regions. We used an axonalSholl analysis to define the architecture of axonal terminals ofipRGCs by arbitrarily assigning the zero point as the center of theventral region of the SCN (Fig. 3D). Therefore, ipRGCs that in-nervate the ventral SCN region will have a peak number of inter-sections from the axonal Sholl analysis closer to 0, whereas ipRGCsthat innervate primarily the dorsal region of the SCN will be furtheraway from 0. A small group of ipRGCs (4 of 20) showed an in-tersection peak lower than 23% of the normalized distance fromthe center, indicating that these cells preferentially targeted theventral SCN. Another group of ipRGCs (7 of 20) had an in-tersection peak of 33–37% of the normalized distance, which wedefined as the medial SCN. The last group of ipRGCs (9 of 20)showed intersection peaks of >41% of the normalized distancefrom the center, which we defined as the dorsal SCN (Fig. 3E).Remarkably, when we overlaid ipRGC axons together, each groupof ipRGC axons seemed to cluster at a particular SCN region (Fig.3 F–I) with a minimum overlap between the dorsal (Fig. 3 H and I)and ventral innervating groups (Fig. 3 F and I). In addition, thesagittal view showed that axonal fields of ventral and medial in-nervating ipRGCs were less spread out compared with dorsal in-nervating ipRGCs (Fig. 3 J–M). We also observed a lack of ipRGCaxons in the most rostral and caudal parts of the SCN (Fig. 3N),similar to previous reports (22). As in uni- and bilaterally projectingipRGCs, the somato-dendritic properties from ipRGCs targetingdifferent areas of the SCN were similar, including total branchpoints, the dendritic surface, total dendritic length, and soma sizes(Fig. S7). Furthermore, the rostral-to-caudal thickness of the axo-nal field from individual ipRGCs seemed to show no obviouschanges (Fig. 3O). Intriguingly, the cell-body location in the retinaof dorsal-SCN–innervating ipRGCs seemed to cluster at thedorsal-temporal region of the retina, whereas ventral and medial-SCN–innervating ipRGCs seemed to cluster at the ventral-nasalregion of the retina (Fig. 3P and Fig. S8). This indicates thepossibility of a crude retinotopic map in the SCN. Therefore, ourresults provide evidence that the retinorecipient region of theSCN is more widespread than originally thought and does notsimply conform to core and shell demarcations.

Retinal Projections to the SCN: Identification of Postsynaptic CellTypes. The widespread projection pattern of ipRGCs in theSCN prompted us to ask whether VIP and GRP cells, comprisingthe core of the SCN, are the only retinorecipient neurons of theSCN. To reveal the retina to SCN synaptic contacts, eyes fromwild-type mice were intravitreally injected with the cholera toxinβ-subunit (CTβ), and coronal SCN sections were colabeled withVIP, GRP, or AVP antibodies (Fig. 4 A–D). VIP-immunopositive(VIP+) cells were densely distributed in the ventral part of the SCNfrom the rostral to the caudal axis (Fig. 4 A and E and Fig. S9), aspreviously shown (15). A confocal z-stack analysis using triple la-beling for VIP, a presynaptic marker (synaptophysin), and retinalaxonal CTβ fluorescence revealed potential synaptic contacts be-tween retinal axons and VIP+ neurons (Fig. 4E). Only a few retinalsynaptic contacts were observed in the soma of VIP+ cells (Fig. 4F).Retinal contacts provided only a small portion of the synaptic inputto VIP+ cells, as nonretinal synaptic contacts (VIP-synaptophysincolocalization) were much more abundant (Fig. 4M). Interestingly,a high number of retinal synaptic contacts was observed on pro-cesses of VIP+ cells (Fig. 4G). A similar analysis was performedusing anti-GRP antibodies. GRP-immunopositive (GRP+) cells inthe medial area of the SCN were sparse (Fig. 4 C and D), and themajority of GRP+ cells received retinal synaptic contacts (Fig. 4 Kand L). As in the case of VIP+ cells, nonretinal synaptic contacts ofGRP were higher than retinal inputs (Fig. 4M). Interestingly, VIP+

cells, which were classically considered to be the principal retinaltarget, received a significantly lower number of retinal contacts atthe soma level compared with GRP+ cells (Fig. 4M). Un-fortunately, we were unable to trace the processes of GRP+ neu-rons to compare them to the retinal synaptic input to VIP+neurons.

Previously, it was assumed that AVP neurons do not receivesynaptic input from the retina (22). Using serial sectioning andSCN reconstruction, we found, as previously reported (22),AVP-immunopositive (AVP+) cells to be located in two distinctshell regions: an outer shell region that is nonretinorecipient andan internal shell region that receives retinal input (Fig. 4H andFig. S9). In the internal shell region, the majority of AVP+ cellsreceive direct synaptic input from retinal axons (Fig. 4 I and J), assuggested by our single ipRGC projection to the dorsal region ofthe SCN (Figs. 2B and 3C). As in the case of VIP+ cells, non-retinal synaptic contacts of AVP+ cells were higher than retinalinputs (Fig. 4M). Importantly, as in the case of GRP+ cells,AVP+ cells received higher synaptic contacts on the soma than

A

B

C

D

E1/2 1/2

1/4

AP stainingcoronal

Reconstructioncoronal sagittal

3v

ox

SCN

3v

ox

SCN

3v

ox

SCN

0 15 20 25 30 35 40 45 50 55

0

1

2

3

Normalized distance from ventral center (%)

Num

bero

fcel

ls

-0.20 -0.15 -0.10 -0.05 0.00 0.05 0.10 0.15 0.200.0

0.1

0.2

0.3

0.4

0.5

0.6

Nor

mal

ized

thic

knes

s

rostral caudalNormalized position in the SCN

-0.7

-0.6

-0.5

-0.4

-0.3

-0.2

-0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

0

5

10

15

20

Normalized position in the SCN

Num

bero

fce l

ls

rostral caudal

N

O

F

G

H

I

J

K

L

M

rostral caudal

3V

3V

3V

3V

coronal sagittal

P D

V

N T

dorsal

ventralmedial

Fig. 3. A single ipRGC preferentially innervates a specific region of the SCN.(A–C) Representative images of single ipRGCs that innervate the ventral (A),medial (B), or the dorsal (C) SCN region. (Scale bar: 50 μm.) (D) Scheme of theaxonal Sholl analysis with the center in the ventral region of the SCN. (E) Theintersection peak distribution of the axonal Sholl analysis from the ventralregion of the SCN. (F–M) Superimposed images of axon fibers of ventrallytargeting (F and J), medially targeting (G and K), dorsally targeting (H andL), and total (I and M) ipRGC projections (F–H and J–L merged by maximumintensity, I and M merged by summation). (N) Accumulation plot of ipRGCaxon fields in the SCN in sagittal view. Dashed lines indicate the rostral andcaudal boundaries of the SCN. (O) Relative sagittal position of each ipRGCaxonal field center and their thicknesses in the SCN (in N and O, position wasnormalized to the total SCN thickness). (P) Spatial distribution of cell bodiesfrom ventral (green), medial (blue), and dorsal (red) SCN-targeting ipRGCs.ox, optic chiasm; 3v, third ventricle.

Fernandez et al. PNAS | May 24, 2016 | vol. 113 | no. 21 | 6049

NEU

ROSC

IENCE

did VIP+ cells. Together, our data showed that all three SCNcell types received input from the retina and that VIP+ cells mayreceive the most input to their processes instead of their somas.

Genetic Labeling of ipRGC Synaptic Contacts in the SCN. CTβ injectionsprovided unexpected retinal synaptic contacts with a population ofAVP+ neurons. To confirm these results using a genetic method,we generated an Opn4CreERT2/+;ROSASynaptophysin–tdTomato mouse linethat expresses a synaptophysin–tdTomato fusion protein exclusivelyin ipRGC axonal terminals (Fig. S1). The use of this line with themelanopsin conditional Cre line allowed us to label only presynapticterminals originating from ipRGCs. For the success of these ex-periments, it is essential to label the majority of ipRGCs, and hencewe injected a high dose of tamoxifen that led to the labeling ofhundreds of ipRGCs in the retina. This synaptophysin–tdTomatoprotein was shown to localize to presynaptic terminals (27) and wasconfirmed here by colocalizing tdTomato with Vglut2 staining (Fig.S10 A and B). We observed a dense pattern of presynaptic retinalterminals across the entire SCN (Fig. S10F), similar to our previousanalysis using single ipRGC tracings and CTβ injections. The pre-synaptic localization of tdTomato expression was evaluated by takingSCN sections from Opn4CreERT2/+;ROSASynaptophysin–tdTomato mice

and coimmunostaining them with the glutamatergic marker Vglut2(Fig. S10 A and B). We found that the great majority of tdTomatopuncta (Fig. S10B, white circles) colocalized with Vglut2 puncta[Mander’s coefficient (M1) = 0.785 ± 0.055; mean ± SEM; n = 4mice]. ipRGC-SCN glutamatergic synapses were identified as pre-synaptic colocalization of tdTomato and Vglut2-positive puncta thatwere in close apposition with postsynaptic AMPA GluR4 staining(Fig. S10 A and B, white boxes). GluR4 is expressed in the SCN;however, not all SCN neurons express GluR4 (28). Therefore,some of the tdTomato and Vglut2 double-positive puncta did notshow apposition with GluR4. Regardless, we observed a clearapposition of tdTomato and Vglut2 double-positive puncta withGluR4 staining, indicating that synaptophysin–tdTomato ex-pressed in ipRGCs made synaptic contact with SCN neurons. Todetermine which SCN neurons receive input from ipRGCs usingthe synaptophysin–tdTomato method, we evaluated tdTomatofocal expression with VIP, GRP, or AVP immunostaining. Wefound synaptic contacts with all three subtypes of SCN neurons(Fig. S10 C–E). Quantification of the number of synaptic contactsper cell (VIP+ cells = 5.50 ± 1.93; GRP+ cells = 8.20 ± 2.95;AVP+ cells = 7.43 ± 1.69; mean ± SEM; n = 5 cells per group)revealed a similar innervation pattern compared with CTβ injec-tions, confirming that ipRGCs innervate all different cell types inthe SCN. In addition, we found that the medial SCN area receiveddense innervation that was significantly higher than those of theventral and dorsal innervation (Fig. S10 F and G), as observedwith single ipRGC tracings.

Synaptic Input from ipRGCs to the SCN Targets a Restricted CellDomain. Because ipRGCs bilaterally innervate the SCN, we eval-uated whether single SCN neurons received input from both eyesand whether the input was localized to specific domains of SCNneurons. We injected CTβ tagged with different fluorophores (488and 594 nm) into each eye, and SCN sections were immunostainedwith anti-synaptophysin, anti-VIP, or anti-AVP antibodies. We foundthat for both VIP+ and AVP+ cells, most SCN neurons receivedinnervation from both eyes (Fig. 5A). Using a 3D reconstruction ofneurons, we measured the distribution of synaptic contacts per celland the spatial relationship between them (Fig. 5 B and C). Wefound a significant association between synaptic contacts originatingfrom both eyes, and in most cases, synaptic contacts from RGCaxons from the ipsi- and contralateral eyes were in close proximity(Fig. 5D). This was in contrast to the diffuse location of synapsesoriginating from nonretinal input (Fig. 5D). This suggests that retinalinput to the SCN neurons has spatially restricted domains.

Monocular and Binocular Light Stimulations Induce Distinct c-FosActivation Patterns in the SCN. Based on the innervation patternto the SCN from both eyes, we decided to evaluate the functionalimplications of such inputs. Induction of the immediate-earlygene, c-Fos, after a light pulse applied during the active phase ofrodents is a commonly used method to evaluate neuronal activa-tion (29). We presented a light pulse (1,000 lx at CT14) to controland monocular-deprived mice and measured induction of c-Fos inthe SCN. Remarkably, in monocular-deprived mice, cFos wasequally induced in both hemispheres of the SCN consistent withour morphological analyses that the SCN receives equal inputsfrom both eyes (Fig. 5 E–I). However, in monocular-deprived mice,cFos levels were significantly reduced compared with those ofcontrol mice (Fig. 5 E–I). In addition, c-Fos induction was observedthroughout the SCN in control animals, whereas in monocular-deprived mice, its induction was localized mainly to the ventral partof the SCN (Fig. 5 E–H, J, and K). These functional data in con-junction with the spatially restricted input of SCN neurons provideevidence that the dorsal part of the SCN possibly requires syner-gistic inputs from both eyes to activate SCN neurons.

DiscussionMultiple Brain Nuclei Are Innervated by a Single M1 ipRGC. In this study,we were able to evaluate the axonal structure of a single ipRGC inthe SCN and compare axonal fields to other retinorecipient areas

Fig. 4. Retinal postsynaptic targets in the SCN. (A–D) Quantification andtopographic distributions of vasoactive intestinal polypeptide (VIP+), argi-nine vasopressin (AVP+), and gastrin-releasing peptide (GRP+) cells in theSCN. (E and F) A representative confocal image showing VIP+ cells in-nervated by retinal axons. A colocalization analysis for VIP (cell-type marker),synaptophysin (presynaptic marker), and CTβ (retinal axonal marker) wasapplied to determine a synaptic contact (triple-colocalization point). Only afew retinal synaptic contacts were observed in the somas of VIP+ cells. (G) The3D reconstruction of a VIP+ cell showing that most of the retinal contacts wereon the processes of the cell (arrows). (H and I) A representative confocal imageshowing that AVP+ cells are located in the shell of the SCN and have receiveddirect retinal innervation. (J) The 3D reconstruction of a AVP+ cell showing thatmany retinal contacts were on the cell body (arrows). (K and L) Similarly, mostGRP+ cells, which are located in the medial core area, received retinal inputs(arrows). (M) Quantification analysis of the total retinal and nonretinal synapticcontacts per cell type. *P < 0.05 vs. VIP+ cells by Tukey’s test. Data are the mean ±SE (n = 25–30 cells). [Scale bars: 5 μm (F, G, and I–L); 25 μm (E and H).]

6050 | www.pnas.org/cgi/doi/10.1073/pnas.1523629113 Fernandez et al.

that received collateral input from the same ipRGC. Using thisconditional approach, we were able to trace M1 ipRGCs, which areexclusively labeled probably due to their high levels of expression ofmelanopsin (9). Consistent with earlier reports (30, 31), we foundthat ipRGCs that innervate the SCN send collateral projections tothe IGL. However, the complexity of the terminal fields was muchless pronounced in the IGL compared with the SCN. In addition, weobserved that ipRGCs also had collateral axons that targeted otherbrain nuclei such as the vLGN, which is part of the lateral geniculatecomplex for visual function; the VMH, which is involved in appetiteand body weight control; the habenular complex implicated in thelimbic system; the olivary pretectal nucleus, which controls the pu-pillary light reflex; the SC, which is important for visual-motor co-ordination (11); and the pretectum/SC region, which was shown tobe involved in sleep regulation (32). Similar to IGL innervation, wealso found that ipRGC axonal terminal fields in these regions wereless elaborate, compared with the SCN. It is important to note thatsome conventional RGCs projecting to the LGN also send collat-erals to the midbrain, including the pretectum and SC, with differentaxonal arbors in each region (33, 34). Interestingly, we observed thatup to five different brain targets received input from a single ipRGC.These results indicate that a single retinal cell can affect brain areasinvolved in many distinct light-mediated behaviors.

The Major Peptidergic Neurons of the SCN Receive Retinal Input.VIP+ cells were classically described as the principal retinal tar-get in the SCN (19, 20). About 60% of VIP+ cells show c-Fos

induction after a light pulse stimulation (21). VIP+ cells are lo-cated in the ventral part of the SCN, in apposition with the opticchiasm, making these cells the best candidates for receiving lightinput. VIP+ cells send projections to the entire SCN, affecting thefunctions of other VIP+ cells and clock cells located in the shell ofthe SCN (15). Here, we showed that in mice ipRGCs form directsynaptic connections with VIP cells, but also with somas of AVPand GRP neurons. These results are the first demonstration, toour knowledge, of a direct retinal input to AVP cells, suggestingthat ipRGCs are poised to provide direct light inputs to severalpeptidergic neuronal populations in the SCN.Consistent with the innervation of several neuronal populations

in the SCN, our single-cell labeling revealed that ∼20% ofipRGCs innervate the ventral part of the SCN, whereas ∼80%innervate the medio-dorsal region. This result was confirmedwith the synaptophysin–tdTomato-inducible system in which welabeled the majority of ipRGC presynaptic terminals in theSCN, indicating that the medio-dorsal SCN region receives thestrongest retinal input.Previous studies indicated that light-induced c-Fos expression

is located primarily in the ventral part of the SCN (35). Becauseour morphological data indicated that neurons in the medio-dorsal SCN, such as AVP and GRP, receive direct retinal inputas strongly as VIP cells, we wondered why previous studies hadshown stronger cFos induction in VIP cells. One possibility isthat the threshold for induction in the dorsal region of the SCNrequires stimulation from both eyes. Our results support thishypothesis, as we observed only c-Fos induction in the dorsalpart of the SCN under binocular but not monocular stimulation.This, in conjunction with the spatially restricted pattern of in-nervation of SCN neurons from the retina, indicates possibleconvergent activation of AVP neurons by ipRGCs. This could bea built-in mechanism whereby VIP neurons can be activated at lowlight levels whereas AVP neurons require higher light intensitiesto stimulate the majority of ipRGCs in the retina due to their lowphoton catch (36).

Atypical ipRGC Ipsilateral Projection Patterns to the SCN. The mouseretina is divided into four quadrants, based on nasal-temporaland dorso-ventral axes. It was shown that RGCs that projectipsilaterally are located in the ventro-temporal quadrant of theretina (25, 37). The results presented here indicate that ipsilat-erally projecting ipRGCs were also located in the ventro-tem-poral part of the retina, although our low number of cells (two intotal) precluded us from drawing a stronger conclusion. In themouse, it was shown that RGCs project to visual targets with a9:1 contralateral:ipsilateral ratio (25, 38). ipRGCs with ipsilat-eral projections (unilaterally) or with contralateral projectionsfollow the same contralateral: ipsilateral ratio as conventionalRGCs. Therefore, contralateral and ipsilateral ipRGCs coulduse Ist2 and Zic2 programs similar to regular RGCs for pathfinding at the midline in the optic chiasm during development(37). However, uniquely to ipRGCs, a subset (66.7%) of con-tralateral ipRGC axons cross the midline and bilaterally in-nervate the SCN. Thus, this bilateral innervation equalizes thefinal innervation volume from one eye to both sides of the SCN,despite the 9:1 contralateral-to-ipsilateral cell number. However,this causes a conundrum of how the contralaterally projectingipRGCs can recross the midline. Although detailed mechanisticunderstanding is currently lacking for retinal ganglion cells, it iswell established that commissure axons that cross the midline donot recross back due to activation of robo1-slit-repulsive signals(39). We propose that the targeting of the ipsilateral SCN by thecontralateral ipRGCs occurs later in development when therobo1-slit or analogous repulsive signals have been down-regu-lated. Consistent with this idea, the SCN is fully innervated byretinal fibers only ∼10 d after birth (40).We also observed that the dorsal SCN-targeting ipRGCs were

clustered in the dorsal-temporal region of the retina, whereasventral and medial SCN-targeting ipRGCs were clustered at theventral-nasal region of the retina. This pattern suggests a crude

Fig. 5. Bilateral retinal innervation to the SCN. A representative confocalimage (A) and 3D reconstruction (B) showing VIP cell receiving innervationfrom both retinas (red CTβ from right eye and green CTβ from left eye).(C) The diagram for distribution analysis. (D) Retinal inputs displayed aparticular pattern of innervation with both retinal synaptic contacts in closeapposition. n = 15–20 cells, **P < 0.01, by Student’s t test. (E and F) c-Fosimmunostaining of the SCN from mouse received no light (E) or a light pulse(1,000 lx) during the active phase (CT 14) (F). (G and H) In monocular-deprived mice, a similar light pulse stimulus still induced a sustained numberof c-Fos+ cells. (I) Quantification of c-Fos+ cell number in the SCN from E–H.No significant differences were observed between the ipsilateral (IpsiL) andcontralateral (ContraL) sides of the SCN in monocular-deprived mice. n = 5;**P < 0.001 vs. control-LP and aP < 0.001 vs. monocular-deprived mice (IpsiLand ContraL) by Tukey’s test. (J) The diagram for topographic distributionanalysis. (K) Quantification of the c-Fos(+) cells topographic distribution inthe SCN from control mice and the IpsiL or ContraL SCN regions frommonocular-deprived mice. *P < 0.05 and **P < 0.001 vs. the control byTukey’s test. [Scale bars: 5 μm (A and B); 100 μm (E).] Data are the mean ± SE.

Fernandez et al. PNAS | May 24, 2016 | vol. 113 | no. 21 | 6051

NEU

ROSC

IENCE

topographic map of ipRGC innervation in the SCN for non-image–forming functions, similar to regular RGCs for image-forming vi-sual functions (41). Because RGCs use the Ephrin/Eph gradient toestablish topographic maps between the retina and LGN/SC,ipRGCs could also use the same Eph receptor to guide them tospecific regions in the SCN. Under environmental conditions, thesun provides overhead illumination, and therefore the ventralretina should receive the highest light input. Thus, ipRGCs locatedin the ventral retina could be the primary ipRGCs that project toVIP and GRP neurons in the ventral SCN for circadian photo-entrainment in the wild.In this study, we provide the first evidence, to our knowledge,

highlighting the elaborate axonal fields of ipRGC innervationspecifically to the SCN and show how the SCN is unique in re-ceiving equal innervation from both eyes compared with otherretinorecipient regions. We also uncovered that AVP neurons do

receive direct retinal input and that equal innervation from botheyes may be essential for activating AVP neurons in response toenvironmental light input.

Materials and MethodsAll animals were handled in accordance with guidelines of the Animal Care andUse Committees of Johns Hopkins University and were approved by the In-stitutional Animal Care and Use Committees of National Taiwan University.Additional materials and methods are described in SI Materials and Methods.

ACKNOWLEDGMENTS. We thank the Technology Commons, College of LifeScience at National Taiwan University for technical assistance with single-cellreconstruction and Johns Hopkins University Mouse Tri-Lab for support anddiscussion. This work was supported by generous contributions from thePEW Charitable Trusts (D.C.F.), National Institutes of Health Grant GM076430(to S.H.), and Taiwan Ministry of Science and Technology Grant MOST 101-2311-B-002-023-MY2 (to S.-K.C.).

1. DeCoursey PJ (1986) Circadian photoentrainment: Parameters of phase delaying.J Biol Rhythms 1(3):171–186.

2. Hattar S, et al. (2003) Melanopsin and rod-cone photoreceptive systems account forall major accessory visual functions in mice. Nature 424(6944):76–81.

3. Wässle H (2004) Parallel processing in the mammalian retina. Nat Rev Neurosci 5(10):747–757.

4. Morin LP (2013) Neuroanatomy of the extended circadian rhythm system. Exp Neurol243:4–20.

5. Hattar S, Liao HW, Takao M, Berson DM, Yau KW (2002) Melanopsin-containingretinal ganglion cells: Architecture, projections, and intrinsic photosensitivity. Science295(5557):1065–1070.

6. Berson DM, Dunn FA, Takao M (2002) Phototransduction by retinal ganglion cells thatset the circadian clock. Science 295(5557):1070–1073.

7. Provencio I, et al. (2000) A novel human opsin in the inner retina. J Neurosci 20(2):600–605.

8. Schmidt TM, Chen SK, Hattar S (2011) Intrinsically photosensitive retinal ganglioncells: Many subtypes, diverse functions. Trends Neurosci 34(11):572–580.

9. Ecker JL, et al. (2010) Melanopsin-expressing retinal ganglion-cell photoreceptors:Cellular diversity and role in pattern vision. Neuron 67(1):49–60.

10. Güler AD, et al. (2008) Melanopsin cells are the principal conduits for rod-cone inputto non-image-forming vision. Nature 453(7191):102–105.

11. Hattar S, et al. (2006) Central projections of melanopsin-expressing retinal ganglioncells in the mouse. J Comp Neurol 497(3):326–349.

12. Baver SB, Pickard GE, Sollars PJ, Pickard GE (2008) Two types of melanopsin retinalganglion cell differentially innervate the hypothalamic suprachiasmatic nucleus andthe olivary pretectal nucleus. Eur J Neurosci 27(7):1763–1770.

13. Welsh DK, Takahashi JS, Kay SA (2010) Suprachiasmatic nucleus: Cell autonomy andnetwork properties. Annu Rev Physiol 72:551–577.

14. Karatsoreos IN, Yan L, LeSauter J, Silver R (2004) Phenotype matters: Identification oflight-responsive cells in the mouse suprachiasmatic nucleus. J Neurosci 24(1):68–75.

15. Kalamatianos T, Kalló I, Piggins HD, Coen CW (2004) Expression of VIP and/or PACAPreceptor mRNA in peptide synthesizing cells within the suprachiasmatic nucleus ofthe rat and in its efferent target sites. J Comp Neurol 475(1):19–35.

16. Abrahamson EE, Moore RY (2001) Suprachiasmatic nucleus in the mouse: Retinal in-nervation, intrinsic organization and efferent projections. Brain Res 916(1-2):172–191.

17. Yan L, et al. (2007) Exploring spatiotemporal organization of SCN circuits. Cold SpringHarb Symp Quant Biol 72:527–541.

18. Tanaka M, et al. (1997) Direct retinal projections to GRP neurons in the supra-chiasmatic nucleus of the rat. Neuroreport 8(9-10):2187–2191.

19. Tanaka M, Ichitani Y, Okamura H, Tanaka Y, Ibata Y (1993) The direct retinal pro-jection to VIP neuronal elements in the rat SCN. Brain Res Bull 31(6):637–640.

20. Ibata Y, et al. (1989) Vasoactive intestinal peptide (VIP)-like immunoreactive neuronslocated in the rat suprachiasmatic nucleus receive a direct retinal projection. NeurosciLett 97(1-2):1–5.

21. Aïoun J, Chambille I, Peytevin J, Martinet L (1998) Neurons containing gastrin-releasingpeptide and vasoactive intestinal polypeptide are involved in the reception of the photicsignal in the suprachiasmatic nucleus of the Syrian hamster: An immunocytochemicalultrastructural study. Cell Tissue Res 291(2):239–253.

22. Lokshin M, LeSauter J, Silver R (2015) Selective distribution of retinal input to mouseSCN revealed in analysis of sagittal sections. J Biol Rhythms 30(3):251–257.

23. Morin LP, Studholme KM (2014) Retinofugal projections in the mouse. J Comp Neurol522(16):3733–3753.

24. Morin LP, Allen CN (2006) The circadian visual system, 2005. Brain Res Brain Res Rev51(1):1–60.

25. Rice DS, Williams RW, Goldowitz D (1995) Genetic control of retinal projections ininbred strains of albino mice. J Comp Neurol 354(3):459–469.

26. Pickard GE (1982) The afferent connections of the suprachiasmatic nucleus of thegolden hamster with emphasis on the retinohypothalamic projection. J Comp Neurol211(1):65–83.

27. Kim J, Matney CJ, Blankenship A, Hestrin S, Brown SP (2014) Layer 6 corticothalamicneurons activate a cortical output layer, layer 5a. J Neurosci 34(29):9656–9664.

28. Mizoro Y, et al. (2010) Activation of AMPA receptors in the suprachiasmatic nucleusphase-shifts the mouse circadian clock in vivo and in vitro. PLoS One 5(6):e10951.

29. Edelstein K, Beaulé C, D’Abramo R, Amir S (2000) Expression profiles of JunB and c-Fosproteins in the rat circadian system. Brain Res 870(1-2):54–65.

30. Pickard GE (1985) Bifurcating axons of retinal ganglion cells terminate in the hypo-thalamic suprachiasmatic nucleus and the intergeniculate leaflet of the thalamus.Neurosci Lett 55(2):211–217.

31. Morin LP, Blanchard JH, Provencio I (2003) Retinal ganglion cell projections to thehamster suprachiasmatic nucleus, intergeniculate leaflet, and visual midbrain: Bi-furcation and melanopsin immunoreactivity. J Comp Neurol 465(3):401–416.

32. Miller AM, Obermeyer WH, Behan M, Benca RM (1998) The superior colliculus-pre-tectum mediates the direct effects of light on sleep. Proc Natl Acad Sci USA 95(15):8957–8962.

33. Tamamaki N, Uhlrich DJ, Sherman SM (1995) Morphology of physiologically identifiedretinal X and Y axons in the cat’s thalamus and midbrain as revealed by intraaxonalinjection of biocytin. J Comp Neurol 354(4):583–607.

34. Bowling DB, Michael CR (1980) Projection patterns of single physiologically charac-terized optic tract fibres in cat. Nature 286(5776):899–902.

35. Meijer JH, Schwartz WJ (2003) In search of the pathways for light-induced pacemakerresetting in the suprachiasmatic nucleus. J Biol Rhythms 18(3):235–249.

36. Do MT, et al. (2009) Photon capture and signalling by melanopsin retinal ganglioncells. Nature 457(7227):281–287.

37. Herrera E, et al. (2003) Zic2 patterns binocular vision by specifying the uncrossedretinal projection. Cell 114(5):545–557.

38. Thompson ID, Morgan JE (1993) The development of retinal ganglion cell decussationpatterns in postnatal pigmented and albino ferrets. Eur J Neurosci 5(4):341–356.

39. Kidd T, et al. (1998) Roundabout controls axon crossing of the CNS midline and defines anovel subfamily of evolutionarily conserved guidance receptors. Cell 92(2):205–215.

40. McNeill DS, et al. (2011) Development of melanopsin-based irradiance detecting cir-cuitry. Neural Dev 6:8.

41. Chen SK, Badea TC, Hattar S (2011) Photoentrainment and pupillary light reflex aremediated by distinct populations of ipRGCs. Nature 476(7358):92–95.

42. Joo HR, Peterson BB, Dacey DM, Hattar S, Chen S-K (2013) Recurrent axon collateralsof intrinsically photosensitive retinal ganglion cells. Vis Neurosci 30(4):175–182.

43. Paxinos G, Franklin KBJ (2013) Paxinos and Franklin’s the Mouse Brain in StereotaxicCoordinates (Elsevier, San Diego), 4th Ed.

44. Hammer Ø, Harper DAT, Ryan PD (2001) PAST: Paleontological statistics softwarepackage for education and data analysis. Palaeontol Electronica 4(1):1–9.

6052 | www.pnas.org/cgi/doi/10.1073/pnas.1523629113 Fernandez et al.